Learning Driver Braking Behavior Using Smartphones, Neural Networks and the Sliding Correlation Coefficient: Road Anomaly Case Study

This paper focuses on the automated learning of driver braking 'signature' in the presence of road anomalies. Our motivation is to improve driver experience using preview information from navigation maps. Smartphones facilitate, due to their unprecedented market penetration, the large-scale deployment of advanced driver assistance systems. On the other hand, it is challenging to exploit smartphone sensor data because of the fewer and lower quality signals, compared to the ones on board. Methods for detecting braking behavior using smartphones exist, however, most of them focus only on harsh events. Additionally, only a few studies correlate longitudinal driving behavior with the road condition. In this paper, a new method, based on deep neural networks and the sliding correlation coefficient, is proposed for the spatio-temporal correlation of road anomalies and driver behavior. A unique deep neural network structure, that requires minimum tuning, is proposed. Extensive field trials were conducted and vehicle motion was recorded using smartphones and a data acquisition system, comprising an inertial measurement unit and differential GPS. The proposed method was validated using the probabilistic Receiver Operating Characteristics method. The method proves to be a robust and flexible tool for self-learning driver behavior.